> ## Content Index
> Fetch the complete content index at: https://blog.aperion.ai/llms.txt
> Use this file to discover other available public pages before exploring further.

# HTTP 402 for AI Agents: Turning Agent Traffic into Paying Customers
- URL: https://blog.aperion.ai/http-402-ai-agents-guide/
- Published: 2026-11-13T17:49:00.000Z
- Updated: 2026-11-17T03:01:37.000Z
- Description: HTTP 402 lets a publisher answer an AI agent with an offer instead of a block. What RFC 9110 says, how x402 and pay per crawl use it, and what to require.
- Author: Craig Alberino
- Tags: Guides, AI Agents, AI Governance, Runtime Governance, #pipeline

HTTP 402 Payment Required lets a publisher answer an AI agent with an offer instead of a block. The response tells the agent, in a form software can read, where the publisher's own checkout is and what the plans are, so the person or company behind the agent can buy access and come back with a credential the publisher issued. Every later call is admitted on that credential and written to a per-call ledger.

Our [publishers guide](https://aperion.ai/blog/publisher-ai-agent-traffic-guide/) sets out the four callers at the front door and the seven steps from measurement to revenue. This guide goes one level down: the 402 response itself, and the credential behind it.

## What HTTP 402 is

RFC 9110, the IETF's HTTP Semantics specification published in June 2022, defines the status code in one sentence: "The 402 (Payment Required) status code is reserved for future use." The standard assigns the number and leaves its use open. Nothing in HTTP says what a 402 body contains or how payment happens.

The neighboring code is settled. RFC 9110 says 403 Forbidden "indicates that the server understood the request but refuses to fulfill it." A 403 ends the exchange. A 402 says the content exists and is available on terms. For a publisher facing agent traffic, 402 is the status code that can carry an offer.

## How the market is using 402 now

Four efforts shape what an agent may see when it reaches a paid resource.

- **x402.** An open payment standard that runs inside ordinary HTTP requests. Its specification, maintained by the x402 Foundation, describes the flow: the resource server responds with `402 Payment Required` and a `PAYMENT-REQUIRED` header carrying the payment requirements, the client builds a payment payload, retries with it, and the server verifies and settles before returning the resource. The Linux Foundation announced the operational launch of the x402 Foundation on July 14, 2026.
- **Pay per crawl.** Cloudflare introduced pay per crawl in private beta on July 1, 2025\. A crawler that requests a priced URL receives a 402 with a `crawler-price` header and can retry with `crawler-exact-price`, or declare `crawler-max-price` up front. The publisher sets a flat, per-request price across its site. Cloudflare's announcement states that "Cloudflare acts as the Merchant of Record for pay per crawl" and that it "aggregates all the events, charges the crawler, and distributes the earnings to the publisher."
- **Web Bot Auth.** The IETF formed the Web Bot Auth working group on October 23, 2025\. Its charter says the group "will standardize methods for cryptographically authenticating automated clients" and provide information about their operators. Pay per crawl uses this approach: crawlers sign requests with HTTP Message Signatures (RFC 9421) using a published key.
- **RSL.** Really Simple Licensing 1.0 became an official specification on December 10, 2025\. It lets a publisher state machine-readable license terms for AI use, including usage categories such as "ai-all," "ai-input" and "ai-index."

x402 covers how a request pays. Pay per crawl covers how a publisher sets a price for signed crawlers and Cloudflare collects it. Web Bot Auth covers which operator runs a client. RSL covers the license terms. A publisher can run more than one.

## Block or allow is the wrong choice

Bot management products give a publisher two settings for agent traffic. A block protects the content and ends the relationship. An allow keeps the relationship and gives the content away. Neither produces a customer.

The third choice is an offer. The agent arrives, presents no credential the publisher recognizes, and receives a 402 that points to the publisher's own checkout. A person at the company behind the agent buys a pass on that checkout, the publisher issues a credential, and the agent returns with it. The publisher remains merchant of record, and the customer record lands in the publisher's own systems.

Three rules keep the offer safe for the business the publisher already has:

- **Humans and subscribers never see it.** A 402 goes to automated callers. A human reader, identified by session state or human verification, and any signed-in subscriber session get the page they always got.
- **Search stays whole.** Verified search crawlers and the publisher's own clients and partners are excluded from offers.
- **Edge responses change after a gate.** Run first in a logging mode that changes no response. Set an approval condition, and change responses after it is met.

## Operator identity is not buyer identity

Signed agent identity tells a publisher who runs the agent. It does not tell the publisher who pays for the read.

The Web Bot Auth charter draws that line itself. It lists "Authenticating the end user of a participating client or agent" as out of scope. A valid signature proves that a request came from a named operator's infrastructure. One operator serves many people and companies, and the signature is the same for all of them.

The gap widens with intermediaries. A search or retrieval service that fetches content for many customers presents its own identity on every request. The buyer behind it is hidden by design.

So admission should rest on a credential the publisher issues. When a request carries that credential, the publisher knows which person or company bought access, under which plan, and which agent is using it. Operator identity is still useful. It tells the publisher who to talk to about a channel arrangement, and it is a good input for ranking demand. It is a poor basis for deciding what to let in.

For intermediaries, the pattern is a pooled credential: a company or a retrieval service receives one under contract, on the condition that each request carries per-customer attribution and is signed.

## What a 402 at the door should carry

The body should be machine-readable and useful to the agent's human.

| Caller presents                                                                   | Response                                                  | What the body carries                                                                                                      | What the ledger records                                      |
| --------------------------------------------------------------------------------- | --------------------------------------------------------- | -------------------------------------------------------------------------------------------------------------------------- | ------------------------------------------------------------ |
| No credential, in a demand cluster chosen for an offer                            | 402 Payment Required                                      | Where to buy, the plans on offer, a configuration snippet a person can paste into an assistant, a reference to the cluster | Payment-required, with the cluster reference                 |
| No credential, not chosen for an offer                                            | Pass through or log, per policy                           | Nothing new                                                                                                                | The outcome, with the cluster reference                      |
| A valid credential within its allotment                                           | 200, full item or excerpt with a pointer, per entitlement | The content                                                                                                                | Licensed or excerpt, the credential, the agent key, the item |
| A valid credential whose allotment is spent or whose plan does not cover the item | Upgrade response with the same structure as the 402       | What to buy to continue                                                                                                    | Payment-required or forbidden, with the credential           |
| A human reader or signed-in subscriber                                            | The normal page                                           | Nothing new                                                                                                                | Not part of the agent flow                                   |

![Decision flow for a request at a publisher's door: a human reader or subscriber gets the normal page; a publisher credential gets the item or an excerpt within its allotment, or an upgrade offer; a caller with no credential gets a 402 with checkout, plans and a setup snippet where an offer was chosen, or passes through and is logged. Every agent outcome goes to one per-call ledger.](https://storage.ghost.io/c/56/dd/56dd2e81-b5ad-4845-bc3a-a53d4e7c40f8/content/images/2026/11/aperion-ip4-at-the-door-2026-11-13.webp)

What an agent gets at the door: an offer, admission on a publisher credential, and one per-call ledger.

A 402 can take one of two pricing approaches. The response can carry no price at all and act as a pointer to the checkout. Or it can carry a deliberately high per-item price, which steers volume buyers toward an account. The cluster reference travels from the 402 into the checkout address, so a purchase can be matched back to the demand that prompted it.

## After the sale: allotments, upgrades and keys

A credential issued at checkout should be of the same class the publisher already uses for its content API. One credential type, one admission rule and one ledger then serve existing subscribers and new agent buyers alike.

Three mechanisms make the credential work for agents. Each is the pattern a publisher should require of any system that admits agents on its behalf.

- **An allotment counted in unique items.** The unit is unique content items per person per month. Repeated retrievals of the same item, headlines and abstracts do not count. Existing subscribers can receive a fair-use allotment. Counting unique items keeps an agent's retries from consuming the plan.
- **An upgrade response when the allotment is spent.** A request that exceeds its allotment, or asks for an item its plan does not cover, receives a response with the same structure as the 402: what to buy, and how. The agent can show its human the next plan instead of failing silently.
- **Individually revocable keys under one pass.** One person's pass issues a separate key to each agent. A lost or misbehaving key is revoked on its own, and the account and the other agents keep working. Every admitted call records which key made it.

Entitlement can vary by plan. A credential with a full-content entitlement receives the item. A credential without one receives an excerpt and a pointer to the full item. The component that checks credentials stores no content. Items stay at the publisher's origin, bulk export is not permitted, and returned items can carry provenance markers.

## The per-call ledger

Every outcome at the door goes to one ledger: the time, the credential, the agent key, the plan, the item, the outcome (licensed, excerpt, payment-required or forbidden) and the cluster reference where there is no credential.

The ledger does three jobs.

- **Invoicing.** Finance bills from the record of admitted calls per credential.
- **Reconciliation.** Payments that arrive through marketplaces and toll networks can be checked against the calls the publisher admitted.
- **Disputes.** When a buyer or a platform questions usage, the ledger answers who read what, when and on which key.

## Turning unidentified demand into a sales file

Most automated traffic will not present a credential at first. That traffic is still demand.

The pattern groups requests that present no publisher credential into clusters by stable request characteristics, and gives each cluster a persistent identifier that stays the same across days. The same entity then runs through the sales record, the offer and the conversion. Clusters are ranked by volume, by the content sections they read, by recency and by estimated commercial value.

Ranked clusters go to the publisher's CRM as records keyed by the cluster identifier, carrying volume, sections read, and first and last seen. A company name is attached when a credential later reveals it, and not before, so the record describes demand without naming a buyer nobody has verified. Sales chooses which clusters receive an offer.

Two rules keep the file honest. A scripted call that succeeds against a login or paywall check is classified as a probe, not as a lost subscription, unless the same client goes on to retrieve content. And clusters that come from an intermediary are flagged, because the buyer behind them is hidden and the path forward is a pooled credential.

When a cluster converts, the purchase is matched to it through the reference carried from the 402, and the cluster is retired from the sales queue and from the offer list. Residual traffic with the same characteristics that still presents no credential is clustered again.

## Questions to ask before you send a 402

- Can your edge tell a human reader or signed-in subscriber from an agent before it sends a 402?
- Does your 402 body tell an agent's human where to buy and how to configure the agent?
- Does checkout run on your own systems, with you as merchant of record?
- Does admission rest on a credential you issue, or on the operator's signature?
- When an allotment runs out, does the agent receive an offer to upgrade?
- Can finance invoice, reconcile and settle a dispute from one per-call record?

## How APERION handles agent traffic

Owned intelligence, governed from the first call. APERION future-proofs the hybrid AI estate: every agent and every model an enterprise rents or owns, governed from one control point inside its own walls.

APERION's Smartflow can front the content API or MCP you already run. In that position, Smartflow can admit a credential you issue, answer other callers with HTTP 402, and write every call to a per-call ledger. The allotments, upgrade responses, revocable agent keys, pooled credentials, excerpts, demand clusters, CRM export and conversion matching described above are the design that builds on that gate. Humans have subscriptions. Agents have APERION Agent Pass.

The accountable human decides, and the record shows who decided, when and under which policy. The [Trust Fabric](https://aperion.ai/blog/trust-fabric/) sets out the four layers underneath: verified identity, access governance, runtime governance, and audit and evidence.

A verified person, the authority that person delegated, the agent that used it, the policy that applied, the decision and the record of all of it. That chain is the product.

## Frequently asked questions

### What is HTTP 402 Payment Required?

HTTP 402 is a status code defined in RFC 9110, the IETF's HTTP Semantics specification, which states that it "is reserved for future use." The standard assigns the number and leaves the format open. Payment protocols such as x402 and edge products such as pay per crawl now use it to tell an automated client that a resource is available on terms, and how to pay or buy access.

### How is x402 different from pay per crawl?

x402 is an open payment standard governed by the x402 Foundation. A server returns a 402 with payment requirements, and the client retries with a payment payload that the server verifies and settles. Pay per crawl is a Cloudflare offering for crawlers that sign their requests: the publisher sets a per-request price, and Cloudflare acts as merchant of record and passes earnings to the publisher.

### Does Web Bot Auth tell a publisher who is paying?

No. Web Bot Auth standardizes cryptographic authentication of automated clients and information about their operators. Its charter lists authenticating the end user of a client or agent as out of scope. A valid signature identifies the company that runs the agent. The person or company the agent reads for, who would pay, is identified by a credential the publisher issues.

### Will human readers or subscribers see a 402?

They should not. A 402 belongs to automated callers that present no credential and fall in demand the publisher chose to make an offer to. Human readers, identified by session state or human verification, and signed-in subscriber sessions receive the normal page. Verified search crawlers are excluded as well, so search visibility and ad measurement are unaffected by the offer.

### What should a 402 response body contain for AI agents?

A useful 402 body is machine-readable and written for the agent's human. It names where to buy, lists the plans, includes a configuration snippet a person can paste into an assistant to present the credential, and carries a reference that links a later purchase back to the demand that prompted it. It can carry no price, or a high per-item price that steers volume buyers to an account.

### How do publishers invoice and settle disputes over agent access?

From a per-call ledger. Each call at the door is recorded with its time, credential, agent key, plan, item and outcome: licensed, excerpt, payment-required or forbidden. Finance invoices from it, reconciles payments received through marketplaces and toll networks against admitted calls, and answers any dispute about who read which item, when and on which key.

## Sources

- IETF, [RFC 9110, HTTP Semantics](https://www.rfc-editor.org/rfc/rfc9110.html#section-15.5.3), June 2022, sections 15.5.3 (402 Payment Required) and 15.5.4 (403 Forbidden). Opened October 31, 2026.
- x402 Foundation, [x402 specification and reference implementation](https://github.com/x402-foundation/x402), "Typical x402 flow." Opened October 31, 2026.
- x402, [x402.org](https://www.x402.org/). Opened October 31, 2026.
- Linux Foundation, [Linux Foundation Announces Operational Launch of x402 Foundation to Standardize Internet-Native Payments for AI Agents and Applications](https://www.linuxfoundation.org/press/linux-foundation-announces-operational-launch-of-x402-foundation-to-standardize-internet-native-payments-for-ai-agents-and-applications), July 14, 2026\. Opened October 31, 2026.
- Cloudflare, [Introducing pay per crawl](https://blog.cloudflare.com/introducing-pay-per-crawl/), July 1, 2025\. Opened October 31, 2026.
- IETF, [Web Bot Auth (webbotauth) working group charter](https://datatracker.ietf.org/wg/webbotauth/about/). Opened October 31, 2026.
- IETF Announce, [WG Action: Formed Web Bot Auth (webbotauth)](https://mailarchive.ietf.org/arch/msg/ietf-announce/4KzSkxT7SqWCtMKcksC5dh5J1Gs/), October 23, 2025\. Opened October 31, 2026.
- RSL Technical Steering Committee, [RSL 1.0 specification announcement](https://rslstandard.org/press/rsl-1-specification-2025), December 10, 2025\. Opened October 31, 2026.

## Related reading

- [How publishers can monetize AI agent traffic](https://aperion.ai/blog/publisher-ai-agent-traffic-guide/)
- [AIDA: agent identity guide](https://aperion.ai/blog/aida-agent-identity-guide/)
- [AI agent governance: MCP, A2A and the new attack surface](https://aperion.ai/blog/ai-agent-governance-guide/)
- [MCP security](https://aperion.ai/glossary-mcp-security/)
- [The enterprise AI gateway guide](https://aperion.ai/blog/enterprise-ai-gateway-guide/)
- [The Trust Fabric](https://aperion.ai/blog/trust-fabric/)